The settings question has a public reference point that small firms can read for free. The joint Cybersecurity Information Sheet "Deploying AI Systems Securely", published on 15 April 2024 by NSA's AI Security Center with CISA, the FBI and the cyber agencies of Australia, Canada, New Zealand and the UK, asks operators to collect logs covering inputs, outputs, intermediate states and errors, and to automate alerts and triggers. For a company with no AI team that is the cost item nobody quotes: five parallel sessions produce five log streams, and somebody has to read them the next morning. The orchestration is the cheap half.
"Worktrees solve a file problem, not a judgment problem" is the line I'll be repeating. I'm running into the same wall exploring this: the sessions scale fine, and my attention to read what comes back is what runs out.
Curious whether you've found a good way to decide which diffs actually need a human read
The settings question has a public reference point that small firms can read for free. The joint Cybersecurity Information Sheet "Deploying AI Systems Securely", published on 15 April 2024 by NSA's AI Security Center with CISA, the FBI and the cyber agencies of Australia, Canada, New Zealand and the UK, asks operators to collect logs covering inputs, outputs, intermediate states and errors, and to automate alerts and triggers. For a company with no AI team that is the cost item nobody quotes: five parallel sessions produce five log streams, and somebody has to read them the next morning. The orchestration is the cheap half.
The shift from individual AI tools to thousands of agents working in parallel could fundamentally change how we think about productivity.
"Worktrees solve a file problem, not a judgment problem" is the line I'll be repeating. I'm running into the same wall exploring this: the sessions scale fine, and my attention to read what comes back is what runs out.
Curious whether you've found a good way to decide which diffs actually need a human read